For text, instead, a popular unsupervised algorithm that can be used to understand a common set of words in a collection of documents is Latent Dirichlet Allocation or LDA.
Note that another algorithm, the Linear Discriminant Analysis, also has the same acronym, but the two algorithms are completely unconnected.
LDA aims to extract sets of homogeneous words, or topics, out of a collection of documents. The math behind the algorithm is very advanced; here we will see just a practical notion of it.
Let's start with an example to explain why LDA is popular and why other unsupervised methods aren't good enough when dealing with text. K-means and DBSCAN, for example, provide a hard decision for each sample, putting ...
